Every builder knows the feeling. Permit on the wall, plans stamped, sub bids sitting in the inbox. And then you stand on the lot at 6:30 in the morning, coffee getting cold, doing math in your head that has nothing to do with concrete or lumber and everything to do with whether the family who said they wanted this house three months ago can still qualify at 7.1% when they signed the contract at 6.6%.
You pull out your phone, check rates, check your lumber account, look at the three other permits you pulled this quarter. Then you drive to the next site where the foundation crew is already set up, because that one you're sure about.
This lot waits.
According to the U.S. Census Bureau's most recent New Residential Construction data, 144,000 single-family homes were authorized for construction but not yet started as of May 2026. That number rose 2.1% in a single month. Each of those permits represents a builder who committed thousands in fees, weeks of plan preparation, and an intent to build, and then stopped at the threshold between paper and dirt, making a decision that amounts to one of the most consequential financial judgments in residential construction with almost no data to support it.
The Cost of Standing Still
Direct fees are the smallest part: a national average of $1,500 to $3,000 for a single-family new construction permit, with plan review adding 65 to 75 percent on top. In Los Angeles, total permit costs for a standard home reach $4,000 or more. In parts of Oregon, a home valued at $368,000 runs $5,370 in permits, plan review, and surcharges before anyone touches a shovel.
Monthly carrying is where the real bleed starts. A finished lot in a typical subdivision ties up $75,000 to $91,000 in capital, per NAHB construction cost data. Finance that at a 60% loan-to-value ratio with a 9% acquisition and development rate, which is close to where AD&C loans sit right now, and you're paying roughly $340 a month in interest alone. Add property taxes on the unimproved parcel, typically $100 to $300 depending on jurisdiction, liability insurance on vacant land, and the opportunity cost of capital parked in dirt instead of deployed on a project that's actually producing revenue, and each idle permitted lot costs a builder somewhere between $600 and $1,000 a month, not counting the risk that the permit itself expires. Most jurisdictions set validity at six to twelve months, with extensions that aren't guaranteed and cost additional fees.
Now multiply.
At $750 per month across 144,000 authorized-but-not-started homes, the industry-wide carrying cost of permit indecision is roughly $108 million per month, or $1.3 billion a year in capital burning while builders wait for a signal that never arrives in a clean, quantified form. That number is approximate, the methodology crude, and not every one of those 144,000 permits represents genuine indecision: some are phased developments, some are build-to-order homes waiting on a buyer's closing, some are strategic holds. But even if only a third represent a builder genuinely stalled between start and wait, that's 48,000 lots and over $430 million a year in carrying costs accruing against homes that exist only on blueprints.
Why Builders Are Freezing
June's data explains the paralysis. Single-family housing starts fell for the third consecutive month in June 2026, slipping 0.2% to a seasonally adjusted annual rate of 895,000 units, according to Reuters. Permits dropped 2.4% to 871,000, the lowest level in ten months, and the NAHB/Wells Fargo Housing Market Index hit 34 in July, its fifteenth straight month below 40, the longest such streak since 2012.
Thirty-seven percent of builders are cutting prices, with the average reduction running about 6%. Sixty-three percent offer incentives. On a home priced at the Census Bureau's January 2026 average of $499,500, a 6% price cut is $29,970 surrendered before the first nail goes in, a concession builders absorb because the alternative is a finished house that sits empty, bleeding carrying costs at triple the rate of an unstarted lot.
Joel Berner, a senior economist at Realtor.com, identified the strategic shift in the May data release: builders are pulling back on spec homes they fear won't sell at full price, choosing instead to market unstarted homes as build-to-order, treating the permit as an option contract, the right to build, held until a committed buyer materializes, with carrying costs functioning as the option premium.
The Information Gap No One Is Filling
Here's what a builder on that lot at 6:30 in the morning actually needs to know: How many competing homes are permitted within two miles and when will they likely start? What's the local absorption rate for new construction in this price band? Are lumber and concrete prices trending up or flattening? Can interested buyers actually qualify at today's rates?
Commercial real estate has had AI-driven answers to these questions for years. TestFit, which raised a $20 million Series A, connects site-fit algorithms to pro forma financial models so a developer sees how design affects economics in real time. Deepblocks interprets zoning overlays and generates feasibility projections for mixed-use projects in minutes. Prophia uses AI to abstract lease data across commercial portfolios, turning hundreds of unstructured documents into queryable intelligence for asset managers.
None of these tools were built for a custom builder doing eight to fifteen homes a year, because the commercial stack assumes a $50 million project with institutional financing and a six-person development team. A residential builder working on a $500,000 spec home outside of Raleigh has a different problem: a single binary decision, start or wait, with a decision window measured in weeks and a total capital exposure that doesn't justify a $2,000-a-month subscription even if one existed.
What the Data-Driven Start Decision Would Require
A residential go/no-go model doesn't need to be TestFit, but it does need five inputs: local permit pull data from the jurisdiction (public record in most counties), MLS absorption rates for new construction in the relevant price band, a materials cost index (NAHB already publishes one monthly), current mortgage qualification thresholds, and the builder's own carrying cost structure. Feed those into a model that updates weekly, and the output is simple: starting today carries an X% probability of selling within 90 days of completion at your target margin, versus holding for Y months at a cost of $Z,000. Not artificial intelligence in any exotic sense — a spreadsheet with live data feeds and a probability engine, the kind of tool a talented analyst could build in a month if someone commissioned it, which no one has.
NAHB's own survey data bears this out: roughly half of single-family builders say they use AI, but only 5% apply it to actual building functions like estimating and scheduling. One in five use it only for marketing and advertising materials, and the gap between AI adoption for brochure copy and AI adoption for a half-million-dollar start-or-wait decision tells you everything about where the industry's priorities sit.
The Counterargument Deserves Its Due
Not everyone thinks those 144,000 idle permits are a problem worth solving with technology. Some experienced builders argue that holding a permit is itself a rational strategy, essentially a cheap option on future demand that keeps the land entitled while waiting for a market signal no algorithm can anticipate, like a sudden rate drop or a competitor pulling out. Others point out that residential markets are so hyperlocal that any predictive model is inherently fragile: the difference between a three-month absorption rate and a nine-month one can come down to a school district boundary rezoning. And production builders will tell you the start decision isn't purely financial but operational. If your framing crew has a gap in three weeks, you start a house to keep them busy, because losing your framer to another builder's pipeline is a six-month setback no spreadsheet captures.
Both objections carry weight, and the hyperlocal one is genuinely difficult: residential markets have smaller sample sizes, more idiosyncratic demand drivers, and less standardized product than commercial, which means any predictive model will be inherently noisier. But noisy prediction still beats no prediction. Commercial real estate made similar arguments about feasibility modeling twenty years ago, and today no institutional developer would break ground on a 200-unit multifamily project without running an AI-assisted pro forma incorporating rent comps, cost indices, and absorption data. That tool didn't eliminate uncertainty; it quantified uncertainty, and that distinction matters.
What This Means If You're Standing on the Lot
If you're a builder holding a permit right now, track your carrying cost per lot per month. Most builders know their hard costs to the penny but couldn't tell you their monthly burn on unstarted inventory without pulling three spreadsheets and a bank statement. If you're building more than ten homes a year, even a crude model would beat the current process: pull weekly permit activity from your county's open data portal (most jurisdictions publish this), cross-reference it with MLS days-on-market for new construction in your price band, and check FRED's weekly mortgage rate data against your lender's current qualification thresholds. That gives you competing supply, absorption rate, and buyer purchasing power — three of the five inputs a real go/no-go model would need, available for free, updated weekly, and currently ignored by most builders in favor of a phone call to their agent and a drive around the subdivision to count for-sale signs.
Residential construction spent $131,734 per home on regulatory compliance in 2026, per the NAHB's June study, a 40% increase in five years. Most of that is mandatory: permit fees, impact fees, code compliance, studies that jurisdictions require before you can touch a shovel. Nobody chose that spending. But the absence of any spending on decision-support tooling for the start/wait question, the single choice that determines whether those mandatory costs lead to a sold home or a carrying-cost drain — that part is a choice, and it's one the industry keeps making by default while builders stand in the dirt, drinking cold coffee, doing math in their heads.
Limitations: The $1.3 billion annual carrying cost estimate uses a simplified per-lot average of $750/month applied uniformly across 144,000 authorized-but-not-started units. Actual carrying costs vary enormously by market, lot value, financing structure, and tax jurisdiction. The Census Bureau's authorized-but-not-started count includes phased developments, build-to-order homes awaiting buyer closings, and strategic holds alongside genuinely stalled projects. This analysis could not determine what fraction of the 144,000 units represents each category. The commercial AI tools cited (TestFit, Deepblocks, Prophia) were not evaluated for accuracy or residential applicability.